Change two frequent GC alloc call sites to use malloc in order to increase perf

from_cpython/Modules/_sre.c

@@ -1180,7 +1180,9 @@ entrance:
                    ctx->pattern[1], ctx->pattern[2]));
 
             /* install new repeat context */
-            ctx->u.rep = (SRE_REPEAT*) PyObject_MALLOC(sizeof(*ctx->u.rep));
+            // Pyston change: use malloc
+            // ctx->u.rep = (SRE_REPEAT*) PyObject_MALLOC(sizeof(*ctx->u.rep));
+            ctx->u.rep = (SRE_REPEAT*) malloc(sizeof(*ctx->u.rep));
             if (!ctx->u.rep) {
                 PyErr_NoMemory();
                 RETURN_FAILURE;
@@ -1194,7 +1196,9 @@ entrance:
             state->ptr = ctx->ptr;
             DO_JUMP(JUMP_REPEAT, jump_repeat, ctx->pattern+ctx->pattern[0]);
             state->repeat = ctx->u.rep->prev;
-            PyObject_FREE(ctx->u.rep);
+            // Pyston change: use malloc
+            // PyObject_FREE(ctx->u.rep);
+            free(ctx->u.rep);
 
             if (ret) {
                 RETURN_ON_ERROR(ret);

from_cpython/Python/dtoa.c

@@ -118,17 +118,18 @@
 
 #include "Python.h"
 
-// Pyston change: disable custom memory managment because it confuses our GC
-#define Py_USING_MEMORY_DEBUGGER 1
-
 /* if PY_NO_SHORT_FLOAT_REPR is defined, then don't even try to compile
    the following code */
 #ifndef PY_NO_SHORT_FLOAT_REPR
 
 #include "float.h"
 
-#define MALLOC PyMem_Malloc
-#define FREE PyMem_Free
+// Pyston change: use normal malloc allocator because it's faster and we can't use the GC
+// because the custom memory managment functions inside this file are not compatible with it.
+// #define MALLOC PyMem_Malloc
+// #define FREE PyMem_Free
+#define MALLOC malloc
+#define FREE free
 
 /* This code should also work for ARM mixed-endian format on little-endian
    machines, where doubles have byte order 45670123 (in increasing address